Related Work

Recent advancements in free-space optical (FSO) communications technology [30] have enabled the establishment of ISLs within LEO satellite networks. Consequently, complete SEC services can be realized within the LEO satellite networks. Several research efforts have explored the potential of SEC technology, particularly in addressing computation challenges for both terrestrial and space missions [2], [27], [31]–[33].

Current Earth observation mission (EOM) advancements facilitate various tasks such as disaster monitoring and resource exploration [8]. Traditional observation scheduling approaches typically focus on capturing multiple RoIs, with each RoI assigned to a single satellite for imaging [6], [9], [10]. Lv et al. [6] presented a holistic solution that precisely models the observation service of a single LEO satellite and allocates several Earth observation tasks to a group of LEO satellites. An iterative algorithm is proposed in [9] to coordinate satellite observation resources to schedule observations for multiple RoIs. However, the above efforts do not allow for the scheduling of fine-grained observations for a large RoI area. In [5], a distributed compression scheme is designed for a large RoI area with high resolution. However, this scheme does not consider the scheduling of observations within the RoI, and each satellite captures a complete shot based on its trajectory.

Various studies have aimed to tackle routing and computation node selection in resource-limited edges for complex inter-task dependent functions. Liu et al. [34] proposed a framework that forms such functions as a service function chain, and the computation node selection for each function is obtained by minimizing the total time cost. In [17], a dependent function embedding algorithm has been designed that considers both routing and computation node selection. However, the congestion in the network is not considered in the above work. Jung et al. [22] proposed a framework that optimizes the routing for DNN inference jobs in distributed networks. Nonetheless, there is still a need to incorporate congestion considerations and explore the potential benefits of raw image splitting for parallel transmission and computation.

近期，自由空间光（FSO）通信技术 [30] 的发展使得在LEO卫星网络中建立星间链路（ISL）成为可能。因此，完整的卫星边缘计算（SEC）服务可以在LEO卫星网络内部实现。已有若干研究工作探索了SEC技术的潜力，特别是在应对地面和太空任务的计算挑战方面 [2], [27], [31]–[33]。

当前，对地观测任务（EOM）的发展推动了灾害监测和资源勘探等多种任务的执行 [8]。传统的观测调度方法通常侧重于捕获多个感兴趣区域（RoI），其中每个RoI被分配给单颗卫星进行成像 [6], [9], [10]。Lv等人 [6] 提出了一种整体解决方案，该方案精确地为单颗LEO卫星的观测服务建模，并将多个对地观测任务分配给一组LEO卫星。文献 [9] 提出了一种迭代算法，通过协调卫星观测资源来为多个RoI进行观测调度。然而，上述工作未能支持对大范围RoI区域进行精细化的观测调度。文献 [5] 为高分辨率的大范围RoI区域设计了一种分布式压缩方案，但是，该方案没有考虑RoI内部的观测调度问题，每颗卫星仅是根据其轨迹进行一次完整的拍摄。

许多研究致力于解决在资源受限的边缘网络中，针对具有复杂任务间依赖关系的功能进行路由和计算节点选择的问题。Liu等人 [34] 提出了一个框架，将这类功能构建为服务功能链，并通过最小化总时间成本来确定每个功能的计算节点选择。文献 [17] 设计了一种同时考虑路由和计算节点选择的依赖性功能嵌入算法。然而，上述工作没有考虑网络中的拥塞问题。Jung等人 [22] 提出了一个框架，用于优化分布式网络中DNN推理作业的路由。尽管如此，将拥塞因素纳入考量，并探索原始图像分割在并行传输与计算方面的潜在优势，仍有待进一步研究。